Oversampling Methods for Handling Imbalance Data in Binary Classification.

Data preparation occupies the majority of data science, about 60–80%. The process of data preparation can produce an accurate output of information to be used in decision making. That is why, in the context of data science, it is so critical. However, in reality, data does not always come in a predefined distribution with parameters, and it can even arrive with an imbalance. Imbalanced data generates a lot of problems, especially in classification. This study employs several oversampling methods in machine learning, i.e., Random Oversampling (ROS), Adaptive Synthetic Sampling (ADASYN), Synthetic Minority Over-sampling Technique (SMOTE), and Borderline-SMOTE (B-SMOTE), to handle imbalanced data in binary classification with Naïve Bayes and Support Vector Machine (SVM). The five methods will be run in the same experimental design and discussed in search of the best and most accurate model for the datasets. The evaluation was assessed based on the confusion matrices with precision, recall, and F1-score calculated for comparison. The AUC and ROC curve is also provided to evaluate the performance of each method via figures. The proposed work reveals that SVM with B-SMOTE has better classification performance, especially in datasets with high similarity characteristics between the minority and majority classes.